Risk aversion

Risk aversion, in formal decision theory, is the preference for a certain outcome over a gamble with the same expected value. The concept originates with Daniel Bernoulli's 1738 paper resolving the St. Petersburg Paradox, formalized by John von Neumann and Oskar Morgenstern (1944) as expected utility maximization with a concave utility function, and given its standard quantitative measures by John W. Pratt (1964) and Kenneth Arrow (1965): the Arrow-Pratt measures of absolute and relative risk aversion.

Risk aversion is distinct from loss aversion, a related but separate concept introduced by Kahneman and Tversky (1979) within prospect theory. Risk aversion operates over total wealth and is captured by curvature of the utility function; loss aversion operates over changes from a reference point and is captured by asymmetric weighting of gains versus losses (losses weighted approximately 2.25x more than equivalent gains). Popular treatments routinely conflate these two concepts. Both describe real psychological phenomena, but they are different formal constructs with different empirical signatures and different implications for how to model decision-making under uncertainty.

The contemporary empirical picture is well-supported in some respects and substantially contested in others. The basic phenomenon — that most people prefer certain outcomes to actuarially equivalent gambles — is robust and replicated across many experimental and field contexts. The Arrow-Pratt formal apparatus is the standard tool in microeconomics, finance, and decision theory. However, Rabin (2000) proved a calibration theorem showing that expected-utility theory with concave utility cannot simultaneously accommodate observed small-stakes risk aversion and plausible large-stakes risk attitudes — if a person rejects a 50-50 lose-$100/gain-$110 gamble at all wealth levels, EU theory implies they must reject implausibly attractive large-stakes gambles. This is widely accepted as a fundamental critique of EU theory as a descriptive framework for real risk attitudes, and it motivates the contemporary preference for reference-dependent models including prospect theory.

Risk aversion matters at three levels with different evidence bases.

For decision theory and economics. Risk aversion is the central concept in the formal theory of decision-making under uncertainty. The Arrow-Pratt apparatus structures most of microeconomic analysis of choice under risk: insurance demand, portfolio choice, asset pricing, optimal contracts under risk-sharing, behavior of firms and consumers. The conceptual framework — preferences over lotteries can be represented by expected utility of outcomes with a concave utility function — has been the dominant model for sixty years, with the von Neumann-Morgenstern axioms providing the foundation that makes the framework mathematically tractable. Most textbook treatment of choice under risk uses this apparatus.

For real-world decision-making. Risk-averse behavior is pervasive in everyday life: buying insurance (paying more than expected loss for certainty), preferring known to ambiguous outcomes, diversifying investments, paying premiums for certainty in contracts. The basic pattern is robust across cultures, contexts, and stakes — though the magnitude varies systematically with population, framing, and stake size. Recognizing risk aversion in your own decision-making (and in counterparties' decision-making) is central to negotiation, contract design, financial planning, and policy analysis.

For the contested theoretical question. The empirical pattern is robust; the question of which formal framework best captures it is contested. The Rabin 2000 calibration theorem showed that EU theory with concave utility cannot describe real risk attitudes consistently across stake sizes. Prospect theory (Kahneman-Tversky 1979) and its successors (cumulative prospect theory, reference-dependent expected utility, rank-dependent utility) provide alternative frameworks with reference-dependent value functions and probability weighting. The contemporary consensus — particularly in behavioral economics — is that risk attitudes are reference-dependent in ways standard EU theory cannot accommodate, but EU theory remains the workhorse for applied analysis because of its mathematical tractability. The distinction between risk aversion (EU framework) and loss aversion (prospect theory framework) is not pedantic: they predict different behavior in different contexts and require different modeling approaches.

The concept of risk aversion has the deepest history of any concept in this glossary, with formalization spanning 1738 to the present.

Bernoulli 1738. Daniel Bernoulli's paper Specimen Theoriae Novae de Mensura Sortis (Exposition of a New Theory on the Measurement of Risk), originally presented to the Imperial Academy of Sciences at St. Petersburg, introduced the foundational insight. Bernoulli was responding to a paradox proposed by his cousin Nicolaus Bernoulli in 1713: imagine a coin-flipping game that pays $2 if heads appears on the first flip, $4 if heads first appears on the second flip, $8 on the third, and so on (doubling each round). The expected value of this game is infinite (sum of 1/2 × 2 + 1/4 × 4 + 1/8 × 8 + ... = 1 + 1 + 1 + ... = infinity). Yet no rational person would pay more than a modest sum to play. This is the St. Petersburg Paradox.

Bernoulli's proposed resolution was the foundational idea: people do not maximize expected monetary value; they maximize expected utility, where utility is a concave function of wealth. Adding $1 to a poor person's wealth produces more utility than adding $1 to a wealthy person's wealth (diminishing marginal utility). Under this assumption, the expected utility of the St. Petersburg game converges to a finite number even though its expected monetary value is infinite. Bernoulli proposed logarithmic utility specifically. The paper was not widely cited in its own time but the English translation by Louise Sommer (with comments by Karl Menger), published in Econometrica in 1954, became standard reference. The Stanford Encyclopedia of Philosophy entry on the St. Petersburg Paradox provides extensive contemporary discussion.

von Neumann-Morgenstern 1944. John von Neumann and Oskar Morgenstern's Theory of Games and Economic Behavior (Princeton University Press) formalized the Bernoullian framework into the modern expected utility apparatus. They proved that if preferences over lotteries satisfy four axioms — completeness, transitivity, continuity, and independence — then preferences can be represented by the expected value of a real-valued utility function. The utility function is unique up to positive affine transformation. Concave utility implies risk aversion in this framework. The vNM theorem is the foundation that makes EU theory tractable; the axioms are now standard in microeconomic theory.

Pratt 1964 and Arrow 1965. John W. Pratt's “Risk aversion in the small and in the large” (Econometrica 32(1):122-136) and Kenneth Arrow's 1965 Yrjö Jahnsson Lectures (later published as Aspects of the Theory of Risk-Bearing, 1965; expanded as Essays in the Theory of Risk-Bearing, 1971) introduced the standard quantitative measures of risk aversion that bear their names. The Arrow-Pratt coefficient of absolute risk aversion is A(w) = -u''(w)/u'(w), the negative ratio of the second to first derivative of utility. The coefficient of relative risk aversion is R(w) = -w · u''(w)/u'(w) = w · A(w). These measures are invariant to positive affine transformations of utility (which represent the same preferences) and capture intuitively meaningful properties: higher A(w) means a person would pay more to insure against a small risk; higher R(w) means a person would devote a smaller share of wealth to risky assets. Specific utility functional forms (CRRA, CARA, mean-variance) correspond to specific patterns of A(w) and R(w). The Pratt-Arrow apparatus is the standard tool in microeconomic analysis of decisions under risk.

Kahneman-Tversky 1979. Daniel Kahneman and Amos Tversky's prospect theory paper (Econometrica 47(2):263-292) introduced a fundamentally different framework that competes with EU theory as a descriptive model of risk attitudes. Three key departures from EU theory: (1) the carrier of value is changes from a reference point, not final wealth; (2) the value function is concave for gains, convex for losses, with a kink at the reference point; (3) probabilities are transformed by a non-linear weighting function that overweights small probabilities and underweights moderate-to-large probabilities. The asymmetric kink at the reference point is loss aversion — the parameter that captures the empirically observed asymmetry between the psychological impact of equivalent gains and losses. Standard estimates of the loss-aversion coefficient (lambda) cluster around 2.0-2.25 — losses hurt approximately 2.25x as much as equivalent gains feel good. Prospect theory is not the same as EU-theory risk aversion; it is a different theoretical framework with different mathematical structure.

Rabin 2000 and the calibration theorem. Matthew Rabin's “Risk Aversion and Expected-Utility Theory: A Calibration Theorem” (Econometrica 68(5):1281-1292) proved a theorem that has become a canonical critique of EU theory as a descriptive framework. The theorem shows: within EU theory, modest risk aversion over small stakes implies absurd risk aversion over large stakes. If a person rejects a 50-50 lose-$100/gain-$110 gamble at all initial wealth levels, the theorem implies they must also reject a 50-50 lose-$1,000/gain-INFINITY gamble — an obviously implausible preference. The mathematical intuition: rejecting a small-stakes gamble with positive expected value requires substantial concavity of utility in a small region; iterating this concavity property across the wealth distribution implies absurd concavity at larger wealth differences. The theorem is widely accepted: it does not prove EU theory wrong, but it shows that EU theory with concave utility cannot describe real risk attitudes consistently. Either people are not approximately EU maximizers, or the small-stakes risk aversion patterns are not real, or both. The Rabin-Thaler 2001 Journal of Economic Perspectives follow-up gave the accessible exposition. The implication: descriptive models of risk attitudes need reference dependence (prospect theory) or other departures from standard EU theory; EU theory remains useful for some applied work but should not be over-interpreted as a description of how people actually decide under risk.

The contemporary state. Risk aversion remains a central concept across economics, finance, and decision theory. The Arrow-Pratt apparatus is the workhorse for applied analysis where mathematical tractability matters. Prospect theory and its successors (cumulative prospect theory by Tversky-Kahneman 1992, rank-dependent utility by Quiggin 1982, reference-dependent expected utility by Koszegi-Rabin 2006) are the dominant frameworks for behavioral analysis of real risk attitudes. The two frameworks are not mutually exclusive: applied work often uses EU-style risk aversion as a first approximation and adds prospect-theory or reference-dependence corrections where the data require them. The honest empirical picture: real risk attitudes are reference-dependent and stake-dependent in ways that the original 1738/1944/1964 EU framework cannot fully accommodate, but EU theory remains tractable and useful for many applications.

Risk aversion has a precise mathematical structure within expected utility theory. Understanding the mechanism matters because the formal structure is what distinguishes risk aversion from loss aversion and from related concepts.

The expected utility framework

A lottery L is a probability distribution over outcomes (typically wealth levels or monetary payoffs). Expected utility theory says: preferences over lotteries can be represented by the expected value of a utility function u(w) applied to outcomes. L is preferred to L' if and only if E[u(L)] > E[u(L')]. The von Neumann-Morgenstern theorem proves this representation exists if and only if preferences satisfy the four vNM axioms (completeness, transitivity, continuity, independence).

Risk aversion as utility concavity

A decision-maker is risk-averse if they prefer the certain expected value of any lottery to the lottery itself. Mathematically: u(E[L]) > E[u(L)] for all non-degenerate lotteries L. By Jensen's inequality, this holds if and only if u is concave (its second derivative is negative). The risk premium π of a lottery is the maximum amount a person would pay to avoid the lottery and receive its expected value instead — a risk-averse person has positive risk premium for any non-trivial lottery.

The Arrow-Pratt measures

Two specific measures, both involving the curvature of the utility function:

• Absolute risk aversion A(w) = -u''(w)/u'(w). The denominator (first derivative, marginal utility) normalizes; the negative of the ratio of curvature to marginal utility. Higher A(w) means more risk aversion at wealth w. This measure is invariant to positive affine transformations of utility (multiplying by a constant or adding a constant gives the same A(w)).
• Relative risk aversion R(w) = -w · u''(w)/u'(w) = w · A(w). Captures risk aversion relative to wealth level — how much a person resists risky variations as a fraction of their wealth.

The Arrow-Pratt approximation gives the risk premium for small risks: for a lottery with mean zero and small variance σ², π ≈ (1/2) · σ² · A(w). The risk premium is approximately half the variance times the absolute risk aversion coefficient. This local approximation is widely used in applied work.

Specific utility functional forms

Three families are particularly common:

• Constant absolute risk aversion (CARA): u(w) = -e^(-aw) for parameter a > 0. The absolute risk aversion coefficient is constant: A(w) = a for all w. CARA utility implies wealth levels do not matter for risky-asset decisions — a millionaire would risk the same dollar amount on a gamble as a person with $1,000 in savings. This is empirically unrealistic but mathematically convenient.
• Constant relative risk aversion (CRRA): u(w) = w^(1-γ)/(1-γ) for parameter γ > 0, with logarithmic utility (u(w) = ln(w)) as the limiting case for γ = 1. The relative risk aversion coefficient is constant: R(w) = γ. CRRA utility implies the share of wealth invested in risky assets is independent of wealth level — richer people invest larger dollar amounts but the same percentage of wealth. This is approximately consistent with observed portfolio behavior and is the standard functional form in macroeconomics and finance.
• Quadratic utility: u(w) = w - bw². Mean-variance preferences (the foundation of modern portfolio theory) require quadratic utility or normally distributed returns. Empirically problematic because absolute risk aversion increases with wealth, but mathematically tractable and central to finance theory.

How EU-theory risk aversion differs from prospect-theory loss aversion

This is the substantive distinction popular treatments miss. Both phenomena involve aversion to losing money, but they have different formal structure:

• Risk aversion (EU framework): defined over total wealth w. The utility function is concave everywhere. A person evaluates lotteries by computing expected utility of final wealth levels. No special role for the status quo or any reference point. Predicts: risk attitudes depend only on final wealth, not on whether the gamble is framed as a gain or loss.
• Loss aversion (prospect theory): defined over changes from a reference point (typically the status quo). The value function is concave for gains, convex for losses, with a kink at the reference point where losses are weighted approximately 2.25x more than gains. Predicts: risk attitudes depend on framing and reference point — the same final-wealth lottery is evaluated differently if framed as gains versus losses.

Empirically, real risk attitudes show both phenomena. The Rabin 2000 calibration theorem demonstrates that EU theory alone cannot accommodate observed small-stakes risk aversion patterns; prospect theory provides reference dependence that does. Standard contemporary modeling in behavioral economics uses prospect theory or reference-dependent expected utility for descriptive work; EU theory remains common for normative and applied work where the mathematical tractability matters more than perfect descriptive accuracy.

Risk aversion is measured through several distinct methodologies, each with strengths and limitations.

Lottery choice tasks (Holt-Laury 2002). The most widely used experimental method. Subjects choose between pairs of binary lotteries with varying probabilities of high versus low outcomes; the pattern of crossover points reveals their risk attitudes. The Holt-Laury task uses ten paired choices and produces a parameter estimate consistent with CRRA utility (the relative risk aversion coefficient γ). Typical estimates from Holt-Laury studies center around γ = 0.3 to 0.5 for general adult populations, with substantial individual variation. This is the dominant experimental measurement tool.

Multiple price list methods. Subjects make a series of choices between a certain amount and a risky lottery, with the certain amount increasing across the list. The switching point reveals the certainty equivalent of the lottery and thus implicit risk aversion. Variations include the Eckel-Grossman method (choosing among lotteries with different expected values and variances) and the Becker-DeGroot-Marschak procedure.

Field measurement via insurance and investment behavior. Some economic research estimates risk aversion from observed real-world choices: insurance deductible choices (Cohen-Einav 2007 used Israeli auto insurance data, finding mean coefficient of absolute risk aversion implying substantial heterogeneity), retirement portfolio allocations (used widely in finance), and game-show contestant behavior (Mulino et al. on Deal or No Deal, Post et al. 2008). Field measurements typically produce risk aversion estimates substantially larger than laboratory measurements — the gap is a methodological puzzle. Rabin and Thaler (2001) argued that any large estimated EU-theory risk aversion is suspect on calibration grounds.

Survey-based measures. Self-report measures (the Dohmen et al. 2011 survey question “How willing are you to take risks?” on a 0-10 scale) have been shown to correlate with experimental measures and with observed behavior in field contexts. Survey measures are cheaper than experimental measures but less precise; they capture general dispositions toward risk rather than specific quantitative risk attitudes.

The fundamental measurement challenge. Different methods produce different estimates of the same person's risk aversion. Laboratory lottery-choice tasks typically imply CRRA coefficients of 0.3-0.5; field insurance choices imply much larger absolute risk aversion; large-stakes decisions imply still other patterns. This is partly the Rabin-calibration problem: EU-theory risk aversion is not a single parameter that describes a person across all decisions, because real risk attitudes are reference-dependent and context-dependent in ways EU theory cannot capture. Studies that report a single risk-aversion coefficient should be interpreted as the parameter that best fits that specific decision context, not a description of the person's “true” risk aversion in some context-independent sense.

What the LBL tools capture. The Crossroads Lab tools (Career Pivot Decision Matrix, Should I Quit My Job?) implement decision-support frameworks where risk attitudes are part of the consideration; they do not produce quantitative risk-aversion parameter estimates. The Cognitive Bias Susceptibility tool in the Behavior Lab measures susceptibility to related decision-making biases including framing effects and loss aversion, which are conceptually adjacent to risk aversion. For users specifically interested in their risk-aversion parameter, established instruments (the Holt-Laury task or Dohmen et al. 2011 single-item measure) remain the published standards.

Risk aversion sits among several closely-related concepts that are routinely confused.

• vs. loss aversion. The most important distinction and the most commonly confused. Risk aversion (Bernoulli/Pratt/Arrow): preference for certain expected value over actuarially equivalent gamble; captured by concave utility over total wealth; defined within EU theory. Loss aversion (Kahneman-Tversky 1979): asymmetric weighting of gains versus losses around a reference point; captured by the kink in the prospect-theory value function; defined within prospect theory. They are different theoretical frameworks with different formal structure. Both describe real psychological phenomena; popular conflation should be resisted. The Rabin 2000 calibration theorem shows that EU-theory risk aversion alone cannot describe real risk attitudes — loss aversion (reference dependence) is needed for a complete picture.
• vs. prospect theory. Prospect theory is the broader framework that includes loss aversion plus probability weighting plus reference dependence. Risk aversion under EU theory is captured by utility concavity over total wealth; prospect theory has its own version (concavity for gains, convexity for losses), but the dominant explanation for risky-choice patterns in prospect theory is the asymmetric loss-aversion kink and the probability-weighting function, not the curvature.
• vs. bounded rationality. Bounded rationality is the broader framework for understanding decision-making under cognitive and environmental constraints. Risk aversion is a property of preferences (how the decider trades off outcomes under uncertainty); bounded rationality is about the cognitive and computational limits on the decider. The two can co-exist: a person can have risk-averse preferences and also be boundedly rational in how they aggregate information about risks. Most empirical work treats risk attitudes and cognitive limits as separate dimensions.
• vs. risk perception. Risk perception is the subjective probability or magnitude a person assigns to uncertain outcomes; risk aversion is how they trade off certainty against expected value given their probabilities. A person can have accurate risk perception and be highly risk-averse, or inaccurate risk perception (over- or underestimating probabilities) with neutral risk attitudes. The two are conceptually separable, though empirically they often co-vary.
• vs. ambiguity aversion. Risk aversion is aversion to outcomes with known probabilities (e.g., a 50-50 coin flip); ambiguity aversion is aversion to outcomes with unknown probabilities (e.g., a gamble where the probability is itself uncertain). Ellsberg (1961) showed these are empirically separable: people prefer known-probability gambles to unknown-probability gambles even when the expected probabilities are the same. Ambiguity aversion has its own formal apparatus (Schmeidler 1989, Klibanoff et al. 2005) distinct from EU theory risk aversion.
• vs. sunk cost fallacy. Sunk-cost reasoning is the tendency to let past unrecoverable costs influence current decisions. Risk aversion is about how to weight uncertain future outcomes. The two phenomena are conceptually distinct: a risk-averse person can still avoid sunk-cost reasoning, and a sunk-cost-prone person can still have neutral risk attitudes. Both contribute to decision quality but through different mechanisms.
• vs. decision fatigue. Decision fatigue is depletion of decision quality across many decisions in sequence. Risk aversion is a stable property of preferences. The two interact: fatigued decision-makers may default to risk-averse choices (e.g., avoiding action), but this is a different mechanism than the underlying risk-averse preference.
• vs. precautionary motive. The precautionary motive is the desire to hold extra wealth as a buffer against future risk. It depends on the third derivative of utility (prudence, Kimball 1990), not just the second derivative (risk aversion). Prudence and risk aversion can move independently — a person can be risk-averse but not prudent, or prudent but not risk-averse. This matters for saving behavior in the presence of uncertain future income.

Example 1 — The insurance decision

A homeowner considers whether to insure her house against fire. The annual premium is $1,200; the probability of total loss in any year is 1 in 1,000; the loss if total destruction occurs is $400,000. The expected loss per year is $400,000 / 1,000 = $400. The insurance premium ($1,200) is three times the expected loss. From an expected-value standpoint, buying insurance is a bad deal: the insurer expects to profit $800 per year on average. Yet most homeowners buy insurance, and the homeowner does too.

This is textbook risk aversion. The certain loss of $1,200 is preferred to the gamble of (a 99.9% chance of zero loss, 0.1% chance of $400,000 loss). The reason is utility concavity: losing $400,000 with 0.1% probability is much more painful in utility terms than losing $1,200 with certainty, because the marginal utility of dollars at the “losing your house” wealth level is enormous compared to the marginal utility at current wealth. The Arrow-Pratt apparatus gives a precise way to estimate the homeowner's risk-aversion parameter from their willingness to pay the premium. If she would pay up to $1,500 but not $2,000 for the insurance, that bounds her absolute risk aversion in the relevant wealth range. Insurance markets exist because of risk aversion: the insurer (who pools risks across many policyholders, so faces close to the expected loss with certainty) profits while the risk-averse homeowner is also better off.

Example 2 — The investment versus job offer case

A 35-year-old engineer is offered a choice. Option A: stay in her current job paying $120,000 per year with 0.5% annual probability of layoff. Option B: join a 50-person startup paying $90,000 base plus equity that, by her own estimate, will be worth $0 with 70% probability, $100,000 with 20% probability, and $1,500,000 with 10% probability over five years. The expected value of the equity is $0 × 0.7 + $100,000 × 0.2 + $1,500,000 × 0.1 = $170,000 over five years, or $34,000/year. The total expected compensation in Option B is $90,000 + $34,000 = $124,000/year — slightly more than Option A's $120,000.

If she were risk-neutral (caring only about expected value), she should pick Option B. Most people in her position pick Option A. The reason is risk aversion combined with reference-dependence: the loss of moving from $120,000 to a 70% chance of $90,000 with zero equity payoff is more painful than the gain of the 10% chance at $1.5M is pleasant. The example also shows the limits of EU-theory analysis: the engineer's decision depends not just on her risk aversion parameter but also on her reference point (the $120,000 current salary anchors what counts as a loss), the probability distribution of the equity payoffs (small probabilities of large payoffs are often underweighted relative to EU theory's predictions), and her psychological framing of the decision (loss aversion makes the downside salient). Pure EU-theory analysis would resolve this with a single concavity parameter; prospect-theory analysis would also use the loss-aversion coefficient and probability-weighting function. Real decisions of this kind reveal both phenomena working simultaneously.

Risk aversion is a foundational concept with substantial empirical support; the substantive limitations are also well-documented.

• The Rabin 2000 calibration theorem is a deep critique of EU theory. Rabin (2000) proved that EU theory with concave utility cannot accommodate real small-stakes risk aversion without implying absurd large-stakes risk aversion. If a person rejects a 50-50 lose-$100/gain-$110 gamble at all wealth levels, they would also reject implausibly attractive large-stakes gambles. The theorem does not prove people are not approximately EU maximizers, but it shows that the standard tool (concave utility over total wealth) is mathematically inadequate as a description of real risk attitudes. The implication: descriptive models of risk attitudes need reference-dependence (prospect theory) or other departures.
• The risk-aversion vs loss-aversion conflation matters. Risk aversion (EU theory: concave utility over total wealth) and loss aversion (prospect theory: kink at reference point) are different formal constructs with different predictions. Popular treatments routinely conflate them, attributing observed behavior to “risk aversion” when prospect theory's loss aversion is the better description. The conflation matters because the two constructs predict different behavior in different contexts — loss aversion predicts framing effects (the same decision is evaluated differently when framed as gains vs losses); EU-theory risk aversion does not.
• Risk attitudes are stake-dependent and context-dependent. Empirical evidence shows people are nearly risk-neutral for small stakes, modestly risk-averse for medium stakes, and substantially risk-averse for large stakes in ways that do not follow a single utility function. A single risk-aversion parameter (CRRA γ, for example) typically cannot fit all stake sizes simultaneously. This is partly the Rabin calibration problem; partly genuine context-dependence in real decision-making.
• Measurement methods produce different estimates. Laboratory lottery-choice tasks (Holt-Laury), survey-based measures (Dohmen et al. 2011), and field insurance choices typically produce different risk-aversion estimates for the same population. The gap between laboratory and field is substantial and has methodological implications: which estimate to trust depends on which decision context matches the application.
• EU theory is normatively well-motivated but descriptively limited. The vNM axioms (completeness, transitivity, continuity, independence) are normatively appealing in many contexts; EU theory provides a coherent framework for normative analysis of decisions under risk. As a descriptive theory of how people actually decide under risk, EU theory is substantially incomplete. The contemporary view: EU theory is appropriate for normative work where you want to characterize ideal decision-making; prospect theory or reference-dependent EU is appropriate for descriptive work where you want to predict actual behavior.
• Substantial individual variation. Risk-aversion parameters vary substantially across individuals on dimensions including age, gender, wealth, education, and cultural context. Population averages can mask important heterogeneity. Studies that report a single risk-aversion estimate for a population should be interpreted as central tendency, not as a description of any specific individual.
• Time and risk preferences may not be separable. Standard EU theory treats risk preferences as separable from time preferences. Empirical work (Andreoni-Sprenger 2012, others) shows the two are often entangled in ways that complicate measurement and modeling. Models that treat risk and time preferences as separate may misestimate both.
• Behavioral departures are not arbitrary; they have structure. Real risk attitudes depart from EU theory in systematic ways: loss aversion, probability weighting, reference dependence, narrow framing. Prospect theory captures several of these systematically; reference-dependent EU theory captures others. The departures are not noise; they are robust regularities that deserve their own theoretical treatment.
• Cultural and demographic variation in risk attitudes is real. Cross-cultural studies show systematic differences in risk attitudes across cultures and across demographic groups within cultures. Studies based on WEIRD (Western, educated, industrialized, rich, democratic) samples may not generalize. Caution is warranted when extrapolating risk-aversion parameters across populations.

The Crossroads Lab tools (Career Pivot Decision Matrix, Should I Quit My Job?) implement decision-support frameworks where risk attitudes are part of the consideration in evaluating choices that involve uncertain outcomes. The Cognitive Bias Susceptibility tool in the Behavior Lab measures susceptibility to related decision-making biases including framing effects, loss aversion, and anchoring — the psychological mechanisms that interact with risk-aversion patterns in real decisions. For users specifically interested in quantifying their risk-aversion parameter, established research instruments (the Holt-Laury lottery task or the Dohmen et al. 2011 single-item measure) remain the published standards. Together these tools provide self-assessment relevant to understanding your own risk-taking patterns in high-stakes life decisions.

Risk aversion is the preference for a certain outcome over a gamble with the same expected value. The concept originates with Daniel Bernoulli's 1738 resolution of the St. Petersburg Paradox via diminishing marginal utility, was formalized by John von Neumann and Oskar Morgenstern (1944) as expected utility maximization with a concave utility function, and was given its standard quantitative measures by John W. Pratt (1964) and Kenneth Arrow (1965) — the Arrow-Pratt measures of absolute and relative risk aversion. Risk aversion is distinct from loss aversion, a related but separate concept introduced by Kahneman and Tversky (1979) within prospect theory. Risk aversion operates over total wealth (captured by utility concavity); loss aversion operates over changes from a reference point (captured by an asymmetric kink in the value function). Popular treatments routinely conflate these constructs. Rabin (2000) proved a calibration theorem showing that EU theory with concave utility cannot accommodate real small-stakes risk aversion without implying absurd large-stakes risk aversion — a fundamental critique that motivates the contemporary preference for reference-dependent models including prospect theory. The basic empirical phenomenon (people prefer certain outcomes to actuarially equivalent gambles) is robust; the Arrow-Pratt apparatus remains the workhorse for applied analysis; descriptively, real risk attitudes are reference-dependent and stake-dependent in ways the original EU framework cannot fully accommodate. The two frameworks (EU theory and prospect theory) are not mutually exclusive in applied work, but the conceptual distinction between risk aversion and loss aversion matters because they predict different behavior in different decision contexts.

This entry is intended as a citable scholarly reference. Choose the format that matches your context. The retrieval date should reflect when you accessed the page, which may differ from the entry's last-reviewed date shown above.

LifeByLogic. (2026). Risk Aversion: Bernoulli, Pratt-Arrow, Rabin. https://lifebylogic.com/glossary/risk-aversion/

LifeByLogic. "Risk Aversion: Bernoulli, Pratt-Arrow, Rabin." LifeByLogic, 14 May 2026, https://lifebylogic.com/glossary/risk-aversion/.

LifeByLogic. 2026. "Risk Aversion: Bernoulli, Pratt-Arrow, Rabin." May 14. https://lifebylogic.com/glossary/risk-aversion/.

@misc{lblriskaversion2026,
  author = {{LifeByLogic}},
  title = {Risk Aversion: Bernoulli, Pratt-Arrow, Rabin},
  year = {2026},
  month = {may},
  publisher = {LifeByLogic},
  url = {https://lifebylogic.com/glossary/risk-aversion/},
  note = {Accessed: 2026-05-14}
}